University of Pennsylvania Working

Papers in Linguistics
Volume 20
Issue 1 Proceedings of the 37th Annual Penn Article 30
Linguistics Conference

1-1-2014

Diachrony or Synchrony? Accounting for the Old

Japanese Particle -tu
Alexander T. Ratte
The Ohio State University, ratte.1@osu.edu

This paper is posted at ScholarlyCommons. http://repository.upenn.edu/pwpl/vol20/iss1/30

For more information, please contact libraryrepository@pobox.upenn.edu.
Diachrony or Synchrony? Accounting for the Old Japanese Particle -tu
Abstract
This paper investigates the Old Japanese (OJ) postposition -tu, a problematic morpheme with both genitive-
like and numeral usages, and proposes a diachronic account of its distribution that eschews the traditional
‘locative-genitive’ analysis in favor of reconstructing pre-OJ *tu as a dual ‘allative / instrumental’ postposition.
I show that both genitive and attributive uses of OJ -tu can be seen as developments out of a pre-OJ allative or
directional sense, and show that reconstructing a pre-OJ allative can account for why OJ -tu is mostly found in
locative expressions. I also propose that the use of OJ -tu in counting could reflect an original instrumental
sense developing via a distributive, functions notably lacking among pre-OJ postpositions. Not only does this
resolve syntactic problems in synchronic genitive analyses attempting to unify these two usages, it also
explains why both bare numeral expressions and numerals suffixed with -tu are attested in Old Japanese.
Reconstructing pre-OJ *tu as a dual ‘allative / instrumental’ reveals a striking match with the Middle Korean
allative / instrumental particle -lwo, from which I reconstruct proto-Korean-Japanese *two via well-attested
sound correspondences between Japanese and Korean. Unlike many of the synchronic analyses proposed thus
far, this diachronic look at OJ -tu offers a unified analysis of its attestations that accounts for its locative
distribution and even reveals a cognate with Korean. At a time when synchronic approaches predominate in
our fields, this analysis helps highlight the value of diachronic approaches towards understanding
distributions.

This working paper is available in University of Pennsylvania Working Papers in Linguistics: http://repository.upenn.edu/pwpl/
vol20/iss1/30
 Diachrony or Synchrony? Accounting for the Old Japanese Particle -tu

Alexander T. Ratte ∗

1 Introduction

This study investigates the Old Japanese (~700 C.E.) particle (postposition) -tu, a morpheme tra-
ditionally labeled “genitive” or “locative genitive” by scholars of Japanese. This particle has per-
plexed historians of the Japanese language. Previous studies such as Martin (1987; 1990) and
Hirata (2001) have sought to understand its unique distribution as extensions of a genitive. How-
ever, proposals for understanding the distribution of -tu synchronically fail to capture important
facts. In this paper, I propose a purely diachronic account as opposed to the traditional, synchronic
genitive analysis, instead reconstructing pre-OJ *tu as an ‘allative / instrumental’ postposition that
can fully explain its usages and distribution in Old Japanese. This diachronic analysis also reveals
a hitherto unrecognized match with the Middle Korean particle -lwo 'allative / instrumental', which
I propose is a cognate morpheme.
Section 2 discusses the usages and distribution of Old Japanese -tu, and addresses some pre-
vious attempts to account for it as a synchronic “genitive”. In Section 3, I introduce my proposal
that pre-OJ *tu was an allative / instrumental particle, and trace the semantic developments lead-
ing to the distribution of -tu attested in Old Japanese. In Section 4, I offer counter-arguments
against the popular theory that Old Japanese -tu is cognate with the Middle Korean genitive parti-
cle -s. In Section 5, I then show how Old Japanese -tu constitutes a phonological and semantic
match with the Middle Korean particle -lwo 'allative / instrumental'. Finally, in Section 6, I sum-
marize the findings of this paper, and conclude with possible avenues for future research.

2 Old Japanese -tu

OJ -tu is found predominantly in clauses involving places or locative expressions; Frellesvig (2010)
refers to OJ -tu as a “locative genitive,” though it may be better termed an “associative”. This
postposition is attested in OJ with a genitive-like sense (Hirata 2001):

1) topo -tu kuni ‘far-off land’ (Manyoshu)

far GEN? land
2) kuni -tu kamwi ‘god of the land’ (Manyoshu)
land GEN? god
3) kami -tu kenwo 'Upper Keno' (Manyoshu)
upper GEN? Keno

However, Hirata (2001) points out that it is also attested in a more complex attributive sense be-
yond a simple genitive:

4) siko -tu wokina ‘a stupid old man’ (Manyoshu)

stupid GEN? old man

A particle -tu in Old Japanese (generally thought to be the same morpheme) is found suffixed onto
numerals post-nominally in counting (in much the same way as a numeral quantifier in Modern
Japanese):

∗
I would like to thank J. Marshall Unger and Brian Joseph for their input and guidance in writing this
paper. I would also like to thank the members of the Changelings group at Ohio State University for com-
ments on an early draft given as a presentation, and my audience at the 37th Penn Linguistics Colloquium. I
am responsible for the contents of this paper and any errors thereof.

U. Penn Working Papers in Linguistics, Volume 20.1, 2014

282 ALEXANDER T. RATTE

5) asi pito -tu ‘one leg’ (Nihon Shoki)

leg one QUANT?
6) u-wo ya-tu kaduk-e ‘dunking 8 cormorants’ (Manyoshu)
cormorant-OBJ eight-QUANT? dunk-INF

However, bare numerals without a suffixed -tu are also attested in Old Japanese to express quanti-
ties of nouns, which means that whatever function -tu carried, its use was not obligatory:

7) puta tose 'two years' (Manyoshu)

two year
8) pito pi pito yo ‘one day and one night’ (Manyoshu)
one day one night
9) mi pe ‘Mie (three layers; place name)’
three layer

Hirata (2001) rightly notes problems in calling OJ -tu a “locative-genitive,” given its use in com-
plex attribution and in counting. Moreover, the productive locative/allative of OJ is -ni; -tu is nev-
er observed as a true locative. However, -tu is not observed as a straightforward possessive either
in Old Japanese, which is also troubling, since we expect genitives to express such constructions.
This makes an analysis of OJ -tu as ‘genitive’ similarly problematic.

2.1 Previous Accounts

The problem of analyzing OJ -tu can be summed up by saying that its distribution in Old Japanese
seems synchronically opaque. Ono et al. (1974) claim that numeral -tu is a numeral classifier, pre-
sumably from OJ -tu ‘genitive,’ but this is a synchronic description, not an explanatory account of
how it came to be, nor does it tell us anything about the relationship of numeral -tu to genitive -tu.
Martin (1987:367) claims that OJ -tu in numeral expressions is a specialized use of GEN -tu, but
once again, this does little more than restate what synchronically could be case (if numeral -tu is
indeed related to genitive -tu), and has little explanatory power to account for how and why OJ -tu
is attested in such a usage. Martin (1990) later claims that -tu in numerals is a kind of nominaliza-
tion identical to GEN -tu, but does not explain how the nominalization relates to counting. Argua-
bly one of the most thorough attempts to explain the uses and distribution of OJ -tu is Hirata
(2001), who claims that OJ -tu derives from a genitive with possessive meaning. Hirata (2001:70)
defends this by arguing that “Possessive is one of the most basic and productive functions of the
genitive, and it is likely to precede complex uses”. Hirata further argues that numerical -tu is a
bound pronominal (Bd-Pro) derived from this genitive -tu. Although his proposals do provide a
framework for understanding how attestations of -tu might be understood in the context of Old
Japanese, Hirata’s account is problematic for three important reasons. First, as mentioned earlier,
OJ -tu is not found in expressions indicating straightforward possession. Second, Hirata himself
notes that attributive usages of OJ -tu would be unexpected, and yet are nevertheless attested in
Old Japanese. Finally, Hirata must assume that the appearance of OJ -tu as a genitive in mostly
spatial relationships is “simply because we do not observe its possessive use, which makes its loc-
ative use salient as a result”. In other words, he provides no explanation for -tu as a “locative-
genitive” other than mere textual happenstance, despite this being the observation that arguably
most demands an explanation. More broadly, Hirata’s (2001) account provides a framework for
understanding the distribution of -tu within the synchronic context of its usage in Old Japanese as
a genitive-like particle, but he rejects the possibility that a synchronic description of -tu in Old
Japanese (as “genitive-like”) may not lead us to the right diachronic account of where usages of
OJ -tu actually come from1. This paper seeks to show that we can better explain the distribution of
OJ -tu as the result of diachronic shift, as opposed to deriving its usages synchronically from a
genitive.

1
Hirata (2001) considers some competing theories that OJ -tu is a reflex of the same morpheme giving
rise to the deictic marker so ‘that,’ but ultimately concludes that -tu must be a genitive.
 DIACHRONY OR SYNCHRONY? ACCOUNTING FOR THE OJ PARTICLE -TU 283

A final word on some previous accounts of -tu. Both Martin (1987; 1990) and Hirata (2001)
assume that OJ -tu must be a cognate morpheme with the Middle Korean particle -s, which is in
fact a true genitive (this morpheme is also known as sai sios or ‘the connecting s’). This assump-
tion figures prominently in their reconstructions of Old Japanese -tu as a true genitive, and has
predisposed some of the literature on -tu towards a genitive account. However, I will show in Sec-
tion 4 that there are reasons for thinking that this is incorrect, and I believe that abandoning this
weak comparison may help to clear up some of the confusion surrounding OJ -tu.

3 Proposal: pre-OJ *tu ‘allative / instrumental’

In this section, I propose that the distribution of OJ -tu can be best explained by reconstructing
pre-OJ *tu as a dual allative / instrumental postposition, which is cognate with the Middle Korean
allative / instrumental postposition -lwo. First, I show how genitive-like and attributive uses of OJ
-tu developed out of a pre-OJ allative particle, and give cross-linguistic evidence for such devel-
opments. Second, I show how uses of OJ -tu with numerals to express counting could have devel-
oped out of a pre-OJ instrumental sense via its use as a numeral distributive, citing similar devel-
opments in English and Russian. I then discuss the advantages of my proposal, and show how this
diachronic reconstruction can resolve many of the problems associated with traditional analyses of
OJ -tu.

3.1 Allative > Associative-Genitive

Instead of seeking to unify the usages of OJ -tu under a single synchronic analysis for Old Japa-
nese, I propose that its distribution can be best understood as the result of diachronic shift, with
genitive-like or associative meanings of -tu originating from a pre-OJ allative particle. Across lan-
guages, we see a relationship between semantic association and directionality. In French, the al-
lative preposition à is used to indicate a possessive sense via association:

10) le livre est à moi

the book is to me
‘the book is to me’ = ‘the book is mine’; (allative) = (possessive by association)

Note also the use of the dative case in Latin to indicate possession and association:

11) mihi nomen

me-DAT name
‘the name to me’ = ‘my name’; (dative) = (possessive by association)

Similarly, we see that in Russian, spatial association through the use of the preposition u ‘at’ is
used to denote personal possession through common spatial association:

12) u menya est’ kniga

at me-GEN2 is book-NOM
‘there is a book at/by me’ = ‘I have a book’; (locative) = (possessive by association)

These possessive, genitive-like expressions developed from directional expressions through asso-
ciation, and this development occurs in many languages across the world. We can generalize a
pathway from direction to a associative or possessive meaning in the following way:

13) N1 to (direction) N2 > N1 (associated) to N2 > N1 associated with N2 > N1 of N2

2
The use of genitive case here is governed by the preposition and unrelated to possession.
284 ALEXANDER T. RATTE

Although both Russian u menya kniga and Latin mihi nomen express a genitive-like meaning, this
meaning is derived from the figurative use of a spatial relationship to associate two nouns. This
gives us a strong basis for arguing that if OJ -tu was a pre-Old Japanese allative particle, it could
have developed into a genitive-like particle that appears to express possession:

2′) kuni-tu kamwi pre-OJ *‘god to the land’ >

‘god (associated) to/with the land’ > ‘god associated with the land’ > OJ ‘god of the land’

This accounts for genitive-like usages of -tu in Old Japanese by showing how the synchronic ap-
pearance of modification need not entail that -tu be derived from a true genitive. Only this account
explains why OJ -tu is a “locative-genitive” (as will be explained in Section 3.5).

3.2 Allative > Attributive

Hirata (2001) claims that the existence of complex attributive interpretations marked with -tu, such
as OJ siko-tu wokina ‘stupid old man,’ are problematic. Here, I show that attributive meanings are
entirely expected in such developments. Some utterances initially expressing a direction cross-
linguistically develop into a more general attributive meaning. For example, a limited set of attrib-
utive expressions in French take the directional/locative preposition à ‘to’:

14) la fille aux cheveux de lin

the girl to the hairs of flaxen

In this example, we can see that cheveux de lin ‘flaxen hairs’ is treated not merely as an associated
physical object, but as a property modifying la fille ‘the girl’, because it is obligatorily preceded
by the definite article les ‘the’ (aux is a fusion of à ‘to’ + les ‘plural definite article’). ‘Flaxen
hairs’ is abstracted as a property, which is then associated with ‘the girl’ by the use of à ‘to’3.
Cheveux de lin ‘flaxen hair’ has come to modify la fille ‘the girl’ in complex attribution, but this
meaning has developed from the use of an allative preposition, and not from a genitive or an at-
tributive. I propose that instances of -tu expressing complex attribution also derive from a pre-OJ
allative sense, with modification arising from figurative association:

4′) siko-tu wokina pre-OJ *‘an old man to stupidity (property)’ >
‘an old man (associated with/to) stupidity (property)’ >
‘an old man associated with stupidity (property)’ > OJ ‘a stupid old man’

Note that the difference between attributive and genitive interpretations of -tu lies in the nature of
the relationship between the two nouns. The modificatory relationship arises from a figurative
association; associating a noun with a quality will naturally give rise to an attributive sense. In this
way, contrary to what Hirata (2001) argues, we expect both attributive and genitive meanings to
arise from association. This gives us a strong basis for thinking that if pre-OJ *-tu is an allative,
we would expect that it would additionally develop into complex attributive expressions.

3.3 Instrumental > Distributive > Counter

I propose that the use of OJ -tu in counting derives not from a genitive, but from a distinct instru-
mental sense (via a distributive) through the following hypothetical developments:

5′) pito-tu ‘one (of something)’

pre-OJ *‘by means of one’ > *‘(counting) by one(s)’ > OJ pito-tu ‘one (of something)'

3
Tellingly, Japanese now uses the copular -no in such expressions: ryokugan no onna ‘the girl of/with
green eyes’, amairo no kami no otome ‘the girl of/with the flaxen hair’.
 DIACHRONY OR SYNCHRONY? ACCOUNTING FOR THE OJ PARTICLE -TU 285

I propose that OJ -tu, when suffixed on numerals, originally expressed a distributive meaning;
to count pito-tu is to count ‘by one/ones’, puta-tu is to count ‘by twos’, etc. This proposal can ex-
plain the syntactic behavior of OJ -tu, namely that it occurs post-nominally and not pre-nominally
as would be expected of a genitive, and accounts for why OJ -tu is only optionally suffixed onto
numerals. Hirata (2001: 71-72) points out that numerals suffixed with OJ -tu “usually occur alone
as pronominals” and “cannot … directly modify nouns as modifiers”. If -tu was really a genitive
particle, it should be able to directly precede a noun as a modifier4. Instead, we can safely account
for syntactic discrepancies between numeral -tu and modification by positing that bare numeral
expressions (such as puta tose ‘two-years’) expressed direct modification within the head noun
phrase, while a numeral suffixed with -tu expressed a distributive sense contained in a separate NP.
Bare numerals could directly modify a noun (just as Korean bare numerals can directly precede
the noun they modify: han salam ‘one-person’ vs. salam han-myeng ‘person-one-QUANT’ ),
whereas the distributive sense highlighted the manner in which the noun was being counted.
Cross-linguistic evidence for the use of an instrumental in counting (via a distributive) can be
found in English, where ‘by’ can be used to indicate numerical distributive, as in ‘counting by
ones,’ ‘one by one,’ and can also be found in Russian, where temporal distributives are expressed
with the instrumental case.
This reconstruction is considerably more speculative, but is supported by several observations.
Old Japanese notably lacks a dedicated instrumental particle; OJ -nite, which is used instrumental-
ly, is clearly an innovation formed from the OJ copular infinitive ni + the gerund suffix -te, mean-
ing 'being' or ‘as’. A transparent innovation in Old Japanese suggests the possibility that an older
instrumental *tu may have been replaced by a copular expression, with only the numeral -tu (now
semantically distinct) surviving this suppletion. In addition, some scholars believe that the closest
hypothesized genetic relatives of Japanese are Korean and/or Tungusic languages such as Manchu
(Unger 2009). Both Korean and Manchu have instrumental particles; if either of these languages
are related to Japanese, it is typologically reasonable to suppose that proto-Japanese may have had
an instrumental particle as well5. Finally, there is the striking comparison of OJ -tu to Middle Ko-
rean -lwo ‘allative / instrumental’. The aforementioned cross-linguistic evidence from English and
Russian shows that instrumentals can express distributives and become used in counting; if the
precursor to Old Japanese did possess an instrumental particle, then it is reasonable to think that its
reflex may lie in a numeral usage. Since we can already reconstruct pre-OJ *tu as ‘allative,’ and
since numeral -tu can be reasonably accounted for by proposing pre-OJ *tu ‘instrumental,’ the
comparison with MK -lwo completes the picture and suggests that such a reconstruction could be
correct.

3.4 Reconstructing Pre-Old Japanese *tu ‘allative / instrumental’

Below, 15) describes the general pathway by which directionality developed into both associative-
genitive and attributive meanings, and 16) shows how instrumentality developed into a quantifier
usage:

15) Allative > Associative-Genitive

N1-tu N2 > ‘N2 to N1’
N1-tu N2 > ‘N2 (associated) to N1’
N1-tu N2 > ‘N2 associated with N1’
N1-tu N2 ‘N2 of N1’

16) Instrumental > Numerical Distributive > Quantifier

4
Hirata’s (2001) proposal that OJ -tu functioned as a bound pronominal is worth considering, but still
does not resolve the problem of why OJ -tu is not used possessively if it was ever a real genitive.
5
The relationship of Japanese to either Korean or Manchu remains in dispute (see Vovin 2010 for a cri-
tique of Japanese-Korean), but Unger (2009) argues that such hypotheses are valid starting points for com-
parative inquiry, and Lee and Ramsey (2011) concur that Korean is likely related to Japanese. The approach
of this paper is not to assume that Japanese and Korean are related, but to show how informed comparative
hypotheses can shed light on opaque internal developments, which in turn support the theory.
286 ALEXANDER T. RATTE

N1 … #-tu > ‘N1 by (#)’

N1... #-tu > ‘counting N1 by (#)’
N1... #-tu ‘(#) of N1’

3.5 Discussion

I have now separately reconstructed pre-OJ *tu as both *'allative' (developing into genitive and
attributive usages) and *'instrumental' (developing into a counting usage via *distributive). I there-
fore propose reconstructing a particle with a dual meaning: pre-OJ *tu 'allative / instrumental'. In
addition to accounting for the relevant data, this proposal accounts for why OJ -tu is not attested as
a basic possessive. It also explains three puzzling observations for OJ -tu.
First, this reconstruction can provide a truly explanatory account for why OJ -tu is used most-
ly in locative expressions. In order for an allative meaning to develop into modification via me-
tonymy, there must be two objects involved that share a spatial relationship: an object that func-
tions as the direction or goal, and an object that undergoes an implied motion towards that direc-
tion or goal. OJ -tu developed into a genitive-like construction only in clauses where two nouns
had a spatial relationship where that spatial relationship resulted in common association. This is
why OJ -tu is found mostly in phrases expressing a location, since locating objects is one of the
most prototypical ways of expressing a spatial relationship.
Second, contrary to Hirata (2001), it is not a problem for this hypothesis that some isolated
cases exist where OJ -tu seems to express not just a spatial association but more general attribution.
We can account for complex attributive relations expressed by OJ -tu by noting how the French
preposition à is used to ascribe properties embodied by one noun onto another noun through asso-
ciation. This example from French demonstrates how an allative/locative particle can assume an
attributive meaning that appears more complex than a spatial association.
Third, this hypothesis can help to explain the idiosyncratic attestations of OJ -tu. I posit that
during the development to Old Japanese, the particle -ni replaces -tu as a means of expressing di-
rectionality, and -tu ceases to be productively used as an allative particle. Allative -ni entirely sup-
plants pre-OJ allative -tu, but does not replace -tu everywhere; we see only those -tu that had de-
veloped a genitive-like or attributive meaning, because all other productive -tu were replaced by -
ni. This helps to explain why we are left with only a limited number of OJ -tu attestations, and
why they all appear to be related to location but are used in genitive and attributive ways.

4 A Critique of Middle Korean GEN -s ~ Old Japanese -tu

As mentioned in Section 2, some scholars have assumed that -tu is cognate with the Middle Kore-
an genitive particle -s, which in turn has led many to believe that OJ -tu should be reconstructed as
a genitive as well. In this section, I seek to show that OJ -tu is a weak cognate match with MK -s,
and that this theory should likely be discarded, which in turn allows us to reconstruct OJ -tu as a
non-genitive postposition. Furthermore, I tentatively propose that MK genitive -s forms a better fit
with the Old Japanese suffix -si, which is an attributive/conclusive marker for adjectives that orig-
inally suffixed onto nominal roots.
Late Middle Korean (c. 15th century, hereafter referred to as Middle Korean or MK) has a
genitive particle that is written as -s. This particle is also sometimes written -t before s or c [ts]:
nwun-t cozo = nwun-s cozo (Martin 1992: 787). Hirata (2001) follows Whitman (1985) in claim-
ing us OJ -tu is related to this MK GEN -s; if this is so, OJ -tu must also be derived from a geni-
tive particle. Hirata (2001) cites two Old Japanese transcriptions of peninsular toponyms in the
Nihonshoki to show how OJ -tu is used similar to how MK GEN -s is used:

17) oko-si tari ‘upper Tari’ ~ OJ kami-tu kenwo ‘Upper Keno’

18) aru-si tari ‘lower Tari’ ~ OJ shimo-tu kenwo ‘Lower Keno’

If -si in okosi, arusi is a genitive suffix, this evidence suggests that MK GEN -s goes back to Old
Korean *si, not *-t. This is problematic, since as Unger (1980) correctly notes, the correspondence
 DIACHRONY OR SYNCHRONY? ACCOUNTING FOR THE OJ PARTICLE -TU 287

of MK s to OJ t in Martin (1966) is actually a correspondence going back to final *s, not medial *s;
in other words, these toponyms attest to a final vowel for the Old Korean form of what shows up
in Middle Korean as GEN -s, which undermines the case that MK GEN -s is cognate with OJ -tu.
Of Martin’s (1966) 8 proposed cognates supporting a correspondence of MK s ~ OJ t, 4 of them
(‘bundle’, ‘dew’, ‘husband’, ‘one’) are dubious; 2 of them (‘hatchet’, ‘mouth’) are in fact also at-
tested with a final -t in Middle Korean (‘hatchet’ nat/nas 낟/낫, ‘mouth’ kwut/kwus 굳/굿) that is
almost certainly the original form; and one of them (‘break’) is not only controverted by Martin’s
other correspondences but is also rejected by Whitman (1985) and Unger (1980). Without more
evidence for this correspondence, it is difficult to justify this comparison.
Finally, the fact that GEN -s appears to be represented as -si with final [i] in these toponyms
suggests that the Old Korean form of Middle Korean GEN -s might be reconstructed as *si. This is
further supported by Old Korean transcriptions of the genitive particle using the phonograms 師
or 叱,, the phonetic interpretations of which are debatable but which both could be reconstructed
with the vowel *i (Ramsey and Lee 2011: 70). This becomes even more problematic for a compar-
ison of MK GEN -s with Japanese -tu, which Hirata (2001) reconstructs as *to, since neither
Whitman (1985) nor Martin (1966) provide correspondences where Korean i matches Japanese o
or u. These are problems with the hypothesis that MK GEN -s is related to OJ -tu.

4.1 Phonetic Stop Epenthesis in Middle Korean

Martin (1992) notes that MK GEN -s had an allomorph -t before s or c, and Hirata offers this as
evidence that GEN -s is cognate with OJ -tu. However, there may be a generalization here that has
escaped Martin and Hirata. In every example that Martin (1992: 787) gives in support of an -s/-t
alternation for GEN -s, the particle is not only followed by -c or -s, but also preceded by -n:

Attested Form Underlying Form

19) nwun t si ’ Gwu ’l ol = nwun s si ‘ Gwu ‘l ol
‘edge of the eyelid’
20) nwun t co ‘zo = nwun s co ‘zo
‘the pupil of the eye’

But there is evidence beyond GEN -s of a similar alternation (Ramsey and Lee 2011):

21) swon-cwo = swon-zwo

‘by hand; personally’

This begs an explanation, and examining other languages for similar trends helps reveal why this
sound change occurs. We see a clear parallel in English, where /s/ has a surface allomorph [ts]
when preceded by a stop:

22) ANSWER: /ænsɚ/ → [æntsɚ]

What is actually happening is the epenthesis of a surface stop following a nasal and before a frica-
tive; in English, [t] after [n]; inserting a voiceless, oral stop bridges the sharp articulatory contrast
between the nasal stop and the voiceless fricative. The parallel here between this epenthesis in
English and the alternation of [s] and [t] in Middle Korean is striking. Middle Korean -t appears
for GEN -s only when it is preceded by the nasal [n] and followed by s or c [ts], both of which
contain the fricative element [s]; this is identical to the environment in which English speakers
epenthesize [t]. The alternation of MK s and t in GEN -s (and other words) is likely due to the
same factors motivating stop epenthesis in English, namely for ease of articulation in complex
clusters. This means that we have a phonetically motivated, cross-linguistic reason for thinking
that this alternation in Middle Korean has nothing to do with an underlying or historical form *t
for GEN -s, but instead reflects a phonetic pronunciation that was coded in Hangul at a time when
288 ALEXANDER T. RATTE

spellings were not yet standardized, and phonemic spellings that characterize modern Hangul was
not yet the norm.

4.2 Proposal: MK GEN -s ~ OJ -si ‘attributive’

I tentatively propose that MK GEN -s goes back to an Old Korean suffix *-si, attested here in the
trasncriptions oko-si and aru-si, a suffix that is cognate not with OJ -tu, but with an Old Japanese
adnominal/conclusive suffix -si. The OJ suffix -si is found in the Early Middle Japanese (Classical
Japanese) adjectival paradigm as a conclusive form, e.g., utuku-si ‘(subj) is beautiful.’ However,
Frellesvig (2010: 83) notes that this same adjectival ending -si is sometimes attested in OJ as an
adnominal (or attributive) form used to directly modify a noun: yo-si nwo ‘Yoshino (the good field;
place name)’. Note that this morpheme -si attaches directly to the root yo- ‘good’; given that Old
Japanese adjectives all appear to be based on nominal roots, we can reconstruct a nominal suffix
of the shape -si that indicated nominal modification. This suffix -si on nouns is thus strongly remi-
niscent of a genitive, and forms a perfect phonological fit with MK genitive -s and possibly Old
Korean *-si.

5 Proto-Korean-Japanese *two ‘allative / instrumental’

In Section 3, I explained and defended my proposal that the distribution of OJ -tu can be explained
by reconstructing a pre-OJ particle *tu with a dual allative / instrumental meaning, and in section 4,
I showed that the claim that OJ -tu is cognate with the Middle Korean genitive -s is weak. I now
propose that pre-OJ allative / instrumental *tu is cognate with the Middle Korean allative / instru-
mental particle -lwo, proto-Korean-Japanese *two. I reconstruct the phonological development of
pKJ *two 'allative / instrumental' as follows:

23) proto-Japanese *two >

*to (loss of distinctive glide *w) >
*tu (Mid-Vowel Raising; Frellesvig, 2010) >
OJ -tu 'genitive/attributive; counter'
proto-Korean *two >
*lwo (consonant lenition; Martin 1996) >
*lo [lo] (semivowel /w/ → labial vowels)
MK -lwo 'allative / instrumental'

The vowel correspondence of OJ u ~ MK wo is supported in both Martin (1966) and Whitman

(1985) and is well attested in other potential cognates. The consonant correspondence between
(non-initial) OJ t ~ MK l is slightly more controversial; it is endorsed by Martin (1966) and reject-
ed by Whitman (1985), but there are Korean-internal reasons that we cannot ignore for thinking
that some Middle Korean /l/ may go back to proto-Korean *d or *t6 (see Martin, 1996, for discus-
sion; he adopts the position that Korean did not have distinctive voicing). In short, the forms pre-
sent a phonological match. Here, I have chosen to reconstruct pKJ *t as the segment whose reflex
is OJ t in non-initial position and MK l in non-initial position by lenition. Reconstructing pre-OJ
*tu as an allative / instrumental particle matches the use of Middle Korean -lwo as both an allative
particle and an instrumental particle:

24) mul -lwo ‘by/with water’

water INSTR
25) nalah-o7 -lwo ‘to the country’ (Worin sokpo; Ramsey and Lee 2011:190)

6
A prime example of why Korean /l/ may descend from proto-Korean *t is the existence of the irregular
T-stems, which alternate between -t- and -l- in their conjugated forms (Martin 1992: 234). That some Middle
Korean /l/ are descended from *t or *d is widely accepted.
7
When preceded by a consonant, a minimal o/u is obligatorily inserted before MK -lwo.
 DIACHRONY OR SYNCHRONY? ACCOUNTING FOR THE OJ PARTICLE -TU 289

country ALLATIVE

Is it problematic to reconstruct an ‘allative / instrumental’ particle with two distinct meanings for
proto-Korean-Japanese? Should we not be searching for a more elementary analysis? I do not
think so. The Uniformitarian hypothesis in historical linguistics means that reconstructed lan-
guages (such as proto-Korean-Japanese) should be typologically no different from the kinds of
languages we see today. Given that linguistic reconstruction almost always involves complex,
creative derivations, there is a somewhat understandable tendency among some linguists towards
reductionism, seeking to understand vast swaths of a language’s grammar in minimalist terms.
This tendency should be tempered with the knowledge that reconstructed languages are also natu-
ral languages, and should adhere to the same kinds of typologies and constraints that we observe
in natural languages today. Clearly, speakers of modern languages tolerate polysemy and morpho-
logical opacity in their lexica, with Korean -lwo ‘allative / instrumental’ being an example of how
words can have separate meanings that need not be analyzed as synchronically derived. In the
same way, there is nothing unreasonable in the idea that speakers of proto-Korean-Japanese would
have done the same; indeed, there is a kind parsimony in a reconstruction that analyzes polysemy
found in the modern language as an inheritance from an ancestor possessing that same polysemy.
In short, while it is conceivable that both allative and instrumental meanings of *two may have
been originally derived from some single (synchronic) form in the language predating proto-
Korean-Japanese (“pre-proto-Korean-Japanese”), no such analysis is necessary to understand how
OJ -tu and MK -lwo could be cognate morphemes. Instead, I believe that reconstructing a pKJ
*allative / instrumental particle is actually a parsimonious theory, as it does no more than project
back what we know is a diachronic fact about Korean.

6 Discussion

In this paper, I have argued that the distribution of OJ -tu is best explained by a purely diachronic
analysis. I do not take the synchronic appearance of OJ -tu as a “genitive” to be any direct indica-
tion of its provenance, and instead show that a reconstruction of pre-OJ *tu ‘allative / instrumental’
best accounts for the distribution of OJ -tu, and reveals a phonological match with Korean -lwo
‘id.’. The idea that Japanese and Korean could be related languages is of course controversial, but
possible Korean-Japanese etymologies can actually help us to uncover important facts about the
Japanese language by suggesting avenues of research that later shed light on internal develop-
ments. This paper has not argued for any reconstruction of OJ -tu that cannot be internally justified,
and the match to Korean -lwo further strengthens this proposal. Indeed, the closeness of the match,
and the fact that both the Japanese and Korean forms can be reconstructed with the same two dis-
tinct meanings, can arguably be said to constitute a particular fact about the two languages (An-
toine Meillet’s fait particulier), a correspondence too unique to be mere coincidence. Future study
may shed more light on the validity of this proposed cognate.
More broadly, this study highlights the importance of diachronic analyses in understanding
synchronically opaque distributions. Many attempts to resolve the problem of -tu have rested on
the assumption that its synchronic appearance in Old Japanese reflects its diachronic origins, that a
synchronic description of -tu is in some way an explanatory account of its distribution. The idea
that a unified account of -tu necessarily involves deriving all of its usages from a “genitive” is
already a theoretical assumption, one that can be deceptive when attempting to understand how
distributions develop. It is only after deconstructing this assumption and de-coupling synchrony
from diachrony that we can consider alternative hypotheses. A single etymon coming to have mul-
tiple reflexes in a synchronic description is not at all surprising, and we do well to avoid equating
a description of synchronic facts with the explanation of how those facts came to be. At a time
when synchronic analyses predominate in our fields, this analysis helps highlight the value of dia-
chronic analyses in understanding distributions.

References
290 ALEXANDER T. RATTE

Frellesvig, B. 2010. A History of the Japanese Language. Cambridge University Press, UK.
Hirata, Y. 2000. Genitive tu in OJ and Historical Changes of Genitive Particles. Japanese/Korean Linguistics,
10.
Hirata, Y. 2001. Genitive particles, historical change, and grammar: issues in Japanese and broader
implications. Doctoral dissertation, Ohio State University.
Martin, S. 1966. Lexical Evidence Relating Japanese to Korean. Language 12:185–251.
Martin, S. 1987. The Japanese Language through Time. New Haven: Yale University Press.
Martin, S. 1990. Morphological clues to the relationships of Japanese and Korean. in Linguistic change and
reconstruction methodology. P. Baldi, ed., Trends in linguistics: studies and monographs 45: 483–
509. Berlin: Mouton de Gruyter.
Martin, S. 1992. A Reference Grammar of Korean. Rutland, Vermont: Charles E. Tuttle Company.
Martin, S. 1996. Consonant Lenition in Korean and the Macro-Altaic Question. University of Hawai’i Press:
Honolulu.
Ono, S., Satake, A. and K. Maeda, eds. 1974. Iwanami Kogo Jiten. Tokyo: Iwanami shoten.
Ramsey, S.R. 1975. Accent and Morphology in Korean Dialects: A Descriptive and Historical Study.
Doctoral dissertation, Yale University.
Ramsey, S.R. 1991. Proto-Korean and the origin of Korean accent. in Studies in the historical phonology of
Asian languages, Boltz and Shapiro eds. Benjamins: Philadelphia.
Ramsey, R. and Lee, K.M. 2011. A History of the Korean Language. Cambridge University Press, UK.
Unger, J.M. 1980. Revision of Proto-Korean-Japanese *s. Korean Linguistics 2: 91–95.
Unger, J.M. 2009. The Role of Contact in the Origins of the Japanese and Korean Languages. University of
Hawaiʻi Press, Honolulu.
Vovin, A. 2010. Koreo-Japonica: A Re-Evaluation of a Common Genetic Origin. Honolulu: University of
Hawaiʻi Press.
Whitman, J. 1985. The Phonological Basis for the Comparison of Japanese and Korean. Doctoral dissertation,
Harvard University.

Department of East Asian Languages and Literatures

The Ohio State University
Columbus, OH 43123
ratte.1@osu.edu